Enhancing credit card fraud detection using traditional and deep learning models with class imbalance mitigation

Introduction: The growing complexity of fraudulent activities presents significant challenges in detecting fraud within financial transactions. Accurate and robust detection methods are essential for minimizing financial losses. Methods: This study evaluates logistic regression, decision tree, and random forest models on real-world credit card datasets, addressing class imbalance and enhancing predictive accuracy. A deep learning model incorporating focal loss was developed to further improve detection performance. The Synthetic Minority Over-Sampling Technique (SMOTE) was applied to mitigate class imbalance, and hyperparameter tuning was conducted to optimize model configurations. Results: Experimental results show that the random forest model achieved the best overall performance, with an accuracy of 99.95%, F1 score of 0.8256, and ROC-AUC of 0.9759. The deep learning model provided the highest precision, demonstrating its potential in minimizing false positives. Discussion: A key novelty of this work is the integration of focal loss within the deep learning framework, enabling the model to focus on hard-to-classify fraudulent transactions. Unlike many prior studies limited to the Kaggle dataset, our approach was validated on both the Kaggle credit card dataset and the PaySim synthetic mobile money dataset, demonstrating robustness and cross-domain generalizability. These findings highlight the effectiveness of combining data preprocessing, resampling techniques, and model optimization for robust fraud detection.